Enhancing Overall Survival Prediction in Head and Neck Cancer with CT Peritumoral Radiomics and Machine Learning

• Main Authors: Salimi, Y., Hajianfar, G., Mansouri, Z., Amini, M., Sanaat, A., Zaidi, H.
• Format: Conference Proceeding
• Language: English
• Subjects: Computed tomography; Feature extraction; Head; Machine learning; Magnetic heads; Prediction algorithms; Predictive models
• ISSN: 2577-0829
• DOI: 10.1109/NSS/MIC/RTSD57108.2024.10658034

Predicting the prognosis after treatment could be beneficial for patient selection and treatment optimization of head and neck malignancies. Machine learning (ML) proved to have the potential to predict overall survival after treatment using imaging textures and radiomics features extracted from gross tumoral volume (GTV). However, medical images from peritumoral tissue may contain information that enhances further the prediction power. The aim of this study was to explore possible improvements in ML algorithms using peritumoral tissue radiomics on a large-scale data. A total number of 2926 head and neck CT images and GTV segmentation pairs from the RadCure dataset (2327 males, and 599 females) were collected. A total of 107 radiomic features, including first order, shape, and textures were extracted from CT images on GTV segmentations and GTVs dilated by a margin varying from 1 to 22 mm with 3 mm steps using PyRadiomics python library. These nine different feature sets were fed to five different models after undergoing six feature selection (FS) methods to predict the overall survival in a ten-fold data split strategy. The model performance was measured in terms of Concordance index (C-Index). The ten-folds overall C-Index averaged over all FS and models were 0.64 \pm 0.023,0.65 \pm 0.02,0.65 \pm 0.03,0.65 \pm 0.03,0.66 \pm 0.03, 0.66 \pm 0.03,0.67 \pm 0.03,0.67 \pm 0.03,0.68 \pm 0.02 for GTV with no dilation, 1,3,4,7,10,13,16,19, and 22 mm dilations respectively. We investigated the added value of peritumoral radiomics in the prediction of overall survival using 30 combinations of FS and models and 9 different inputs. Our results revealed the importance of tissues surrounding GTV CT radiomics in head and neck cancer prognosis prediction using ML algorithms.